Multivariate statistical monitoring of ETo: A new approach for estimation in nearby locations using geographical inputs

The standard equation used to calculate reference evapotranspiration (ETo) requires many parameters that are not available or reliable in most cases. Alternative equations have been developed in the literature relying only on a limited number of climatic records. A different approach is proposed in this work based on exogenous ETo records from locations with similar climatic conditions. Principal components analysis (PCA) is applied to ETo data recorded from 30 weather stations, which allows ETo estimation of past or present missing data when local climatic inputs are not available. This approach resulted more accurate for gap infilling purposes than other tested methods, with average absolute relative errors around 9%. The proposed methodology, which was developed in the 1990s for the monitoring and diagnosis of chemical processes, can only be applied if previous ETo measurements are available. If this is not the case, a new procedure based on principal components regression is proposed to estimate ETo when only local geographical data are available. The resulting models present average absolute relative errors around 10%. The relationships among stations were described by means of a map that can be used for estimation purposes using one or two neighboring stations.

► Multivariate statistical monitoring approach to estimate ETo when no local climatic inputs are available.
► Principal components analysis (PCA) useful tool to characterize similarities among weather stations.
► Four principal components are enough to describe systematic variability of ETo data.
► PC regression approach allows ETo estimations using geographical inputs with AARE around 10%.